[memory-leak] Dispose previous uniform buffer in SKRuntimeEffectUniforms.Reset()

[github-actions]

AI-generated fix — produced by the Fixer - Memory Leak agentic workflow using the memory-leak-fixer skill (focus area 0: Undisposed native handle). The fix is validated by a red→green regression test (proven failing without the fix, passing with it).

The fix

SKRuntimeEffectUniforms.Reset() replaced its owned, native ref-counted SKData buffer with a freshly-allocated one without disposing the previous buffer, leaking one native SkData allocation per call (reclaimed only by the finalizer, non-deterministically). This is the classic undisposed native handle pattern.

The idiomatic fix (matching focus area 0's Fix ✓: dispose the owned handle deterministically when it is replaced) disposes the old buffer before assigning the new one:

public void Reset ()
{
    if (data.Size == 0)
        return;

    var old = data;
    data = SKData.Create (old.Size);
    old.Dispose ();
}

The temporary local avoids the focus area's Watch out pitfalls: no field is nulled before disposal, no borrowed/same-instance handle is disposed, and SKData.Create (→ sk_data_new_uninitialized) genuinely returns a fresh owned handle, so disposing the replaced one is correct — not a double-free.

Proof (red→green)

Added regression test SKRuntimeEffectTest.Generic.ResetDisposesPreviousUniformBuffer, which captures the old buffer's native handle via reflection and asserts the wrapper is deregistered from the handle dictionary immediately after Reset() (deterministic, no GC needed):

# Without the fix (red):
dotnet test tests/SkiaSharp.Tests.Console/SkiaSharp.Tests.Console.csproj \
  -p:TargetFramework=net10.0 -p:TargetFrameworks=net10.0 \
  -- --filter-method "*ResetDisposesPreviousUniformBuffer*"
# => Failed: 1, Passed: 0

# With the fix (green):
# => Failed: 0, Passed: 1

Both directions verified (revert ⇒ red, re-apply ⇒ green). Neighbouring non-GPU SKRuntimeEffectTest tests continue to pass; the 18 GPU failures in that class are pre-existing "Failed to open X display" environmental failures unrelated to this change.

Scope note

• Clear framework bug, not a usage footgun — the leak is entirely inside SkiaSharp-owned managed code.
• Empirically proven, not just statically reasoned.
• ABI-stable: no public signature changed; the change is internal to Reset().
• Managed C# only (binding/SkiaSharp/SKRuntimeEffect.cs + test); no generated files or externals/skia/** touched.

Areas Affected

• SkiaSharp (core bindings)

Fixes #4782

Labels: tenet/performance, perf/memory-leak.

Generated by Fixer - Memory Leak · opus48 · 319.4 AIC · ⌖ 28 AIC · ⊞ 12K · ◷